Event
AdaDex: Training Generalizable Dexterous Skills from a Single Human Hand Trajectory
1 reports · 1 sources
Yuxuan Kuang and other researchers propose AdaDex, a goal-conditioned reinforcement learning framework that converts a few seconds of human hand-object trajectories (from motion capture data or HOI reconstruction from in-the-wild RGB(D) video) into generalizable dexterous skills, covering behaviors such as functional grasping, object repositioning, finger-gait in-hand reorientation, and tossing and catching. Its core Hybrid Dexterous Mimicking trains a shared policy in two sets of large-scale parallel environments: environments with dense action-tracking objectives inject human priors, while environments with noisy initial states and goals and sparse object-success rewards force the skill to work even when deviating from the reference trajectory.
Report timeline
- AdaDex: Learning Generalizable Dexterous Skills from a Single Trajectory via Hybrid Dexterous Mimicking
AdaDex is a goal-conditioned reinforcement learning framework that turns a few seconds of human hand-object trajectory (from motion capture data or HOI reconstruction from in-the-wild RGB(D) video) into generalizable dexterous skills. Its core, Hybrid Dexterous Mimicking, trains a shared policy across two sets of massively parallel environments: environments with dense motion tracking objectives inject the human prior, while environments with noised initial states and goals and a sparse object-success reward force the skill to work even when it deviates from the reference trajectory.
arXiv cs.RO ·